REVIEW 1 major objections 4 minor 283 references
Test-Time Scaling in Reasoning LLMs: Inference Regimes, Evaluation, and Reproducibility
T0 review · 1 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read Test-time scaling results are properties of the full inference protocol, not of a scalar compute budget.
desk verdict Careful framework paper whose core argument holds, but the headline release count is wrong by three orders of magnitude and must be corrected before the corpus claims are credible. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the implicit prefix tree $T(x)$ induced by an autoregressive model $p_\theta$ from prompt $x$, with terminal leaves $L(x)$; every test-time algorithm is viewed as a budgeted procedure $A_B$ that interleaves primitive operations (token generation, prefix expansion, verifier or judge calls, terminal reduction) under a cumulative cost bound $B$, yielding an induced proposal distribution $q_\pi(y|x)$ over completed leaves. The evaluation machinery is the discovery–stability profile $S_{k,t} = \frac{1}{Q}\sum_q \kappa_{k,t}(p_q)$, where $\kappa_{k,t}$ is the binomial tail probability that $k$ fresh attempts yield at least $t$ correct candidates; Pass@k, pass_k, majority-vote consistency, and related metrics are coordinates, functionals, or bounds of this profile. The framework also decomposes protocol cost as $C_{\mathrm{total}} = C_{\mathrm{gen}} + C_{\mathrm{eval}}$ with $C_{\mathrm{eval}} = C_{\mathrm{signal}} + C_{\mathrm{control}} + C_{\mathrm{decision}}$, and an energy–gate decomposition of token decoders defines the proposal distribution induced by truncation and reweighting rules.
What would settle it
For a fixed checkpoint and question set, generate two response banks that differ only in decoder gate order (top-p then top-k versus top-k then top-p) or only in the reducer, then compare accuracy across many prompts under protocol-matched bootstrap intervals; if the protocol changes move reported accuracy by less than the uncertainty in repeated runs, the central claim of protocol dependence is weakened, while systematic shifts beyond the intervals would confirm it.
Extended reading notes
Core claim
The paper's central claim is that test-time scaling should be formalized as a family of budgeted inference algorithms over the implicit prefix tree of an autoregressive model, with performance treated as a property of the complete inference system rather than of model weights or a single scalar budget. It distinguishes three regimes: single-trajectory sequential scaling, where one active path is extended or revised; leaf-level scaling, where a bank of completed candidates is reduced by voting, verifier selection, or minimum Bayes risk; and prefix-level scaling, where search over unfinished states uses continuation-value estimates and controllers such as beam, best-first, or MCTS-style search. The paper then develops evaluation principles: end-to-end system performance $M_B$ and candidate-bank diagnostics are different estimands; the discovery–stability profile $S_{k,t}$ unifies repeated-sampling metrics; and reproducibility is separated into exact replay versus distributional reproducibility, with uncertainty propagated through the actual aggregation and stopping protocol. Empirically, the paper shows on fixed response banks that candidate discovery outpaces selection: median Pass@80 reaches 82% while the all-correct coordinate falls to 15%, and swapping one reducer for another can move a configuration's accuracy from 75.6% down to 65.8% as the bank grows.
Load-bearing premise
The framework assumes that every test-time algorithm can be broken into primitive operations whose costs add up, and that the process gives a stable distribution over completed candidates; if sharing computation between prefixes or stop-when-you-want rules makes costs non-additive or makes the candidates dependent, the quantitative claims need to be reworked.
Editorial extensions
If this is right
- Comparing reasoning models by a scalar budget and an accuracy number alone is insufficient; reports should specify the checkpoint, prompt, decoder, controller or reducer, verifier or judge, budget decomposition, and stopping rule together.
- Common repeated-sampling metrics are not competing measures but coordinates or bounds of one discovery–stability profile, so they can be reported coherently and related to each other.
- Shared-bank comparisons isolate the aggregation stage, end-to-end comparisons measure the deployed system, and compute accounting must include generation, evaluation, control, and decision costs rather than only sample counts.
- Released response banks allow new reducers and verifiers to be evaluated without regenerating candidates, supporting exact replay of the paper's analyses.
- Distributional reproducibility requires uncertainty estimates that propagate through the aggregation and stopping protocol, such as prompt bootstrap intervals computed after the reducer, rather than a single seeded run.
Reading between the lines
- This suggests a practical test: if every leaderboard entry carried a protocol fingerprint (prompt hash, decoder gate order, verifier identity, budget decomposition, stopping rule), many apparent ranking disagreements between studies would likely resolve into protocol differences rather than model differences.
- The profile's discovery–stability shape also invites a two-axis performance comparison: plotting Pass@k against all-correct consistency would make the trade-off between finding a correct answer and finding it reliably the explicit object of comparison, rather than hiding it in a scalar.
- Because the additive-cost premise is the framework's weakest point, a natural extension is to replace token-count budgets with wall-clock budgets under tree-structured inference kernels that share key–value cache; the paper notes such sharing can change the practical compute frontier but does not build it into the formal accounting.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper develops a systematic account of test-time scaling for reasoning LLMs, formalizing it as budgeted inference over the implicit prefix tree of an autoregressive model and distinguishing single-trajectory sequential scaling, leaf-level scaling with terminal reduction, and prefix-level search. It proposes evaluation principles that treat the entire inference system as the evaluated object, introduces a discovery–stability profile that unifies repeated-sampling metrics, distinguishes exact replay from distributional reproducibility, and specifies protocol-matched reporting of compute and uncertainty. The framework is instantiated on MMLU-Pro, BBH, competition mathematics, and SuperGPQA with a large released corpus of reasoning traces, and the empirical sections include shared banks, bootstrap intervals, verifier diagnostics, and explicit caveats about token caps and effort confounding.
Significance. If the claims hold, the paper provides a genuinely useful vocabulary and evaluation protocol for test-time scaling: the three-regime taxonomy, the shared-bank versus end-to-end distinction, and the discovery–stability profile are well designed and fill a real gap in the literature. The empirical study is unusually careful: bootstrap intervals are propagated through the actual aggregation protocol, candidate banks are shared across reducers, token-cap rates and length limits are disclosed, verifier limitations (reference-assisted versus reference-free, ordinal rather than calibrated) are stated, and per-candidate seeds and serialization are specified. These practices are exactly the kind the paper argues for, and the release, once its size is stated correctly, will be a reusable resource. The central argument that reported performance depends on the full inference protocol is well supported by the formalization and the experiments.
major comments (1)
- [Abstract and Section 1 (release-size claim)] The Abstract states “assemble over 2 billion full reasoning traces” and the Introduction states “over 2 million reasoning traces,” but Table 1 and Section 5 report a total of 1,948,821 traces. The only 2-billion-scale number in the paper is the 1,832,922,842 “completion positions” in the compact Parquet snapshot described in Section G.5, which is a count of token positions, not full reasoning traces; the Introduction’s “over 2 million” also exceeds the tabulated total of 1,948,821. Because the release is a principal stated contribution and the paper itself (Section 3.5) argues that artifact counts are reproducibility requirements, this internal inconsistency is load-bearing. Please correct both statements (for example, “1,948,821 traces, including over 1.8 billion completion positions in the signal-rich snapshot”) and verify that every occurrence count in the abstract, introduction, body, and conclusion agrees with the release census.
minor comments (4)
- [Section 3.4 (Beta prior)] The default Beta(1,1) prior for the latent success probabilities p_q is a modeling choice; a short sensitivity check (for example, Beta(2,2) or an empirical-Bayes prior derived from an auxiliary bank) would strengthen the claim that the discovery–stability profile comparisons are not prior-driven.
- [Sections 2.1 and 3.3 (cost accounting)] The decomposition C_total = C_gen + C_eval assumes additive costs; since Section C.3 correctly notes that tree-structured kernels can reuse key–value caches and change the wall-clock frontier, the main text should state explicitly that the decomposition is a reporting convention rather than a physical identity, and that incommensurable costs should be reported separately.
- [Section G.3 (exact replay metadata)] For the broad knowledge and symbolic reasoning block, the dataset-hub revisions are unavailable and no chat template is recorded; these omissions are honestly disclosed, but given Section 3.5’s exact-replay requirements, please mark each release block as supporting exact replay, distributional reproducibility, or both, so readers know which artifacts are sufficient for which claim.
- [Section 4.3 and Table 14 (effort confounding)] The 16.12% token-cap rate for gpt-oss high effort is correctly reported, but the cap rate should appear alongside the main accuracy numbers in the main text rather than only in the appendix, since it directly qualifies the effort comparison and is a model of the disclosure the paper recommends.
Circularity Check
No material circularity: the framework is self-contained and the empirical claims are descriptive; only minor non-load-bearing self-citations appear.
full rationale
The paper's central claims are definitional and empirical, not derived from fitted parameters. The three-regime taxonomy and the prefix-tree formalization in Section 2.1 are explicit modeling choices; the evaluation profile in Section 3.4 is defined as a binomial-tail expectation, and the identities Pass@k=S_{k,1}, pass_k=S_{k,k}, and Maj@k≥S_{k,⌊k/2⌋+1} are algebraic consequences of that definition rather than predictions. The paper labels the profile a candidate-bank diagnostic and repeatedly states that it does not give the success probability of an aggregation rule, so no fitted value is renamed as a result. The empirical study uses fixed response banks and exact without-replacement statistics, with no parameter fitted to benchmark outcomes and then called a prediction. The only self-citations (Hariri et al. 2026a,c) support tie-breaking conventions, ranking uncertainty, and reproducibility framing; the present paper re-derives its framework in full, so these citations are not load-bearing. Acknowledged limitations, including ordinal verifier calibration, the reference-assisted Compass diagnostic, search-induced candidate dependence, and the high-effort token-cap confounding, are honest disclosures rather than circular steps. One correctness concern outside circularity remains: the abstract and contribution list claim 'over 2 billion full reasoning traces' while Table 1 and Section 5 total 1,948,821 traces, and the only 1.8-billion figure in Section G.5 counts completion token positions; this is an internal release-statistic inconsistency, not a definitional reduction.
Assumptions & free parameters
free parameters (3)
- Beta(1,1) prior for latent prompt success probabilities p_q
- Contextual Compass null-adjustment temperature =
1.5
- Pointwise Qwen ordinal score mapping =
A=20 ... T=1
assumptions (5)
- domain assumption An autoregressive language model and a local generation policy induce a well-defined proposal distribution over completed leaves via the implicit prefix tree.
- domain assumption Task utility U_x(y) is a well-defined scalar for benchmarks, either exact correctness or a graded score.
- domain assumption Compute costs are additive and can be split into generation, signal, control, and decision costs.
- domain assumption For the independent-attempt profile, latent per-prompt success probabilities p_q exist and the Beta-Binomial model applies.
- standard math Standard probability identities are accepted: binomial tails, hypergeometric without-replacement estimates, beta-binomial conjugacy, and the MBR equivalence of self-consistency.
invented entities (1)
-
Discovery-stability profile S_k
Cite this review
Pith. "Pith review of Test-Time Scaling in Reasoning LLMs: Inference Regimes, Evaluation, and Reproducibility." pith.science (2026). https://pith.science/paper/GHTU5OQC
@misc{pith2026260804001,
author = {Pith},
title = {Pith review of: Test-Time Scaling in Reasoning LLMs: Inference Regimes, Evaluation, and Reproducibility},
year = {2026},
howpublished = {\url{https://pith.science/paper/GHTU5OQC}},
note = {Machine review of arXiv:2608.04001}
}
read the original abstract
Large language models can solve substantially harder reasoning problems with more inference-time compute. The term "test-time scaling," however, now covers diverse inference algorithms that extend deliberation along a single trajectory, sample completed candidates and aggregate them through voting or verification, or search over unfinished partial states. These algorithms differ in their statistical structure, compute accounting, and failure modes. Treating these procedures as interchangeable under a single scalar "budget," or reporting accuracy without the inference protocol that produced it, makes results difficult to compare across studies. We develop a systematic account of test-time scaling along three axes. First, we formalize test-time scaling as budgeted inference over the implicit prefix tree of an autoregressive model and distinguish three structural regimes: single-trajectory sequential scaling, leaf-level scaling with terminal reduction, and prefix-level scaling. Second, we treat the evaluated object as the entire inference system and develop evaluation principles that separate end-to-end system performance from candidate-bank diagnostics. We introduce an evaluation profile whose coordinates and simple functionals recover or bound common repeated-sampling metrics, and prescribe protocol-matched reporting of compute and uncertainty. Third, we specify reproducibility requirements for inference protocols, distinguishing exact replay from distributional reproducibility and identifying the artifacts needed to support each. We also organize the open-weight reasoning ecosystem by model-side and interface mechanisms, apply these principles to broad-knowledge, symbolic-reasoning, and competition-mathematics benchmarks, and assemble over 2 billion full reasoning traces for release with progressively richer verifier and token-level signals.
Figures
Figures from the paper (5 more)
Reference graph
Works this paper leans on
-
[1]
arXiv preprint arXiv:1904.09751 , year=
The curious case of neural text degeneration , author=. arXiv preprint arXiv:1904.09751 , year=
arXiv 1904
-
[2]
arXiv preprint arXiv:2408.10577 , year=
Optimizing large language model hyperparameters for code generation , author=. arXiv preprint arXiv:2408.10577 , year=
-
[3]
arXiv preprint arXiv:2407.01082 , year=
Turning up the heat: Min-p sampling for creative and coherent llm outputs , author=. arXiv preprint arXiv:2407.01082 , year=
-
[4]
arXiv preprint arXiv:2506.13681 , year=
Turning Down the Heat: A Critical Analysis of Min-p Sampling in Language Models , author=. arXiv preprint arXiv:2506.13681 , year=
-
[5]
arXiv preprint arXiv:2402.06925 , year=
A thorough examination of decoding methods in the era of llms , author=. arXiv preprint arXiv:2402.06925 , year=
-
[6]
arXiv preprint arXiv:2310.01693 , year=
Closing the curious case of neural text degeneration , author=. arXiv preprint arXiv:2310.01693 , year=
-
[8]
arXiv preprint arXiv:2407.21787 , year=
Large language monkeys: Scaling inference compute with repeated sampling , author=. arXiv preprint arXiv:2407.21787 , year=
-
[9]
Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages =
s1: Simple Test-Time Scaling , author =. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages =. 2025 , publisher =
2025
Show all 283 references
-
[10]
B leu: a Method for Automatic Evaluation of Machine Translation
Papineni, Kishore and Roukos, Salim and Ward, Todd and Zhu, Wei-Jing. B leu: a Method for Automatic Evaluation of Machine Translation. Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics. 2002. doi:10.3115/1073083.1073135
2002
-
[11]
arXiv preprint arXiv:2009.10297 , year=
Codebleu: a method for automatic evaluation of code synthesis , author=. arXiv preprint arXiv:2009.10297 , year=
2009 arXiv
-
[12]
Advances in Neural Information Processing Systems , volume=
Spoc: Search-based pseudocode to code , author=. Advances in Neural Information Processing Systems , volume=
-
[15]
arXiv preprint arXiv:2203.11171 , year=
Self-consistency improves chain of thought reasoning in language models , author=. arXiv preprint arXiv:2203.11171 , year=
-
[16]
Advances in neural information processing systems , volume=
Solving quantitative reasoning problems with language models , author=. Advances in neural information processing systems , volume=
- [17]
-
[18]
arXiv preprint arXiv:2411.00640 , year=
Adding error bars to evals: A statistical approach to language model evaluations , author=. arXiv preprint arXiv:2411.00640 , year=
-
[19]
arXiv preprint arXiv:2504.21303 , year=
Confidence in large language model evaluation: A bayesian approach to limited-sample challenges , author=. arXiv preprint arXiv:2504.21303 , year=
-
[20]
Advances in Neural Information Processing Systems , volume=
Benchmarking llms via uncertainty quantification , author=. Advances in Neural Information Processing Systems , volume=
-
[21]
The Fourteenth International Conference on Learning Representations , year=
Don't Pass@k: A Bayesian Framework for Large Language Model Evaluation , author=. The Fourteenth International Conference on Learning Representations , year=
-
[23]
The Annals of Statistics , volume=
Bootstrap Methods: Another Look at the Jackknife , author=. The Annals of Statistics , volume=
-
[24]
Forty-first International Conference on Machine Learning , year=
Chatbot arena: An open platform for evaluating llms by human preference , author=. Forty-first International Conference on Machine Learning , year=
-
[25]
arXiv preprint arXiv:2412.18407 , year=
A statistical framework for ranking llm-based chatbots , author=. arXiv preprint arXiv:2412.18407 , year=
-
[26]
arXiv preprint arXiv:2508.11452 , year=
Inclusion Arena: An Open Platform for Evaluating Large Foundation Models with Real-World Apps , author=. arXiv preprint arXiv:2508.11452 , year=
-
[27]
2nd Workshop on Models of Human Feedback for AI Alignment , year=
Dropping Just a Handful of Preferences Can Change Top Large Language Model Rankings , author=. 2nd Workshop on Models of Human Feedback for AI Alignment , year=
-
[29]
Findings of the Association for Computational Linguistics: NAACL 2025 , pages=
Re-evaluating automatic LLM system ranking for alignment with human preference , author=. Findings of the Association for Computational Linguistics: NAACL 2025 , pages=
2025
-
[30]
Improving Reproducibility in Machine Learning Research (A Report from the
Pineau, Joelle and Vincent-Lamarre, Philippe and Sinha, Koustuv and Larivi. Improving Reproducibility in Machine Learning Research (A Report from the. Journal of Machine Learning Research , volume =. 2021 , url =
2021
-
[32]
Proceedings of Machine Learning and Systems , volume =
Accounting for Variance in Machine Learning Benchmarks , author =. Proceedings of Machine Learning and Systems , volume =. 2021 , url =
2021
-
[34]
Second Conference on Language Modeling , year =
A Sober Look at Progress in Language Model Reasoning: Pitfalls and Paths to Reproducibility , author =. Second Conference on Language Modeling , year =
-
[35]
Yuan, Jiayi and Li, Hao and Ding, Xinheng and Xie, Wenya and Li, Yu-Jhe and Zhao, Wentian and Wan, Kun and Shi, Jing and Hu, Xia and Liu, Zirui , journal =. Give Me. 2025 , url =
2025
-
[36]
arXiv preprint arXiv:2504.04823 , year=
Quantization hurts reasoning? an empirical study on quantized reasoning models , author=. arXiv preprint arXiv:2504.04823 , year=
-
[38]
, booktitle=
Tahmasivand, Ahmad and Zahran, Noureldin and Al-Sayouri, Saba and Fouda, Mohammed and Khasawneh, Khaled N. , booktitle=. LM-Fix: Lightweight Bit-Flip Detection and Rapid Recovery Framework for Language Models , year=
-
[39]
2025 , archivePrefix=
INTERLACE: Interleaved Layer Pruning and Efficient Adaptation in Large Vision-Language Models , author=. 2025 , archivePrefix=
2025
-
[40]
Transactions on Machine Learning Research , year=
Holistic evaluation of language models , author=. Transactions on Machine Learning Research , year=
-
[41]
Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , pages=
Dynabench: Rethinking benchmarking in NLP , author=. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , pages=. 2021 , url=
2021
-
[42]
International Conference on Learning Representations , year=
Proving test set contamination in black-box language models , author=. International Conference on Learning Representations , year=
-
[43]
Nature , volume=
A benchmark of expert-level academic questions to assess AI capabilities , author=. Nature , volume=. 2026 , doi=
2026
-
[44]
International Conference on Machine Learning , year=
DyCodeEval: Dynamic Benchmarking of Reasoning Capabilities in Code Large Language Models Under Data Contamination , author=. International Conference on Machine Learning , year=
-
[45]
International Conference on Learning Representations , year=
LiveBench: A Challenging, Contamination-Limited LLM Benchmark , author=. International Conference on Learning Representations , year=
-
[46]
First Conference on Language Modeling , year=
Gpqa: A graduate-level google-proof q&a benchmark , author=. First Conference on Language Modeling , year=
-
[47]
arXiv preprint arXiv:2311.07911 , year=
Instruction-following evaluation for large language models , author=. arXiv preprint arXiv:2311.07911 , year=
-
[48]
Advances in Neural Information Processing Systems , volume =
Wang, Yubo and Ma, Xueguang and Zhang, Ge and Ni, Yuansheng and Chandra, Abhranil and Guo, Shiguang and Ren, Weiming and Arulraj, Aaran and He, Xuan and Jiang, Ziyan and Li, Tianle and Ku, Max and Wang, Kai and Zhuang, Alex and Fan, Rongqi and Yue, Xiang and Chen, Wenhu , titl...
2024
- [49]
-
[50]
International Conference on Learning Representations , year=
BeyondBench: Contamination-Resistant Evaluation of Reasoning in Language Models , author=. International Conference on Learning Representations , year=
-
[51]
arXiv preprint arXiv:2602.10367 , year=
LiveMedBench: A Contamination-Free Medical Benchmark for LLMs with Automated Rubric Evaluation , author=. arXiv preprint arXiv:2602.10367 , year=. doi:10.48550/arXiv.2602.10367 , url=
-
[52]
Challenging
Suzgun, Mirac and Scales, Nathan and Sch. Challenging. Findings of the Association for Computational Linguistics: ACL 2023 , pages =. 2023 , doi =
2023
-
[53]
arXiv preprint arXiv:2103.03874 , year=
Measuring mathematical problem solving with the math dataset , author=. arXiv preprint arXiv:2103.03874 , year=
-
[54]
2025 , archivePrefix=
Simulating a Bias Mitigation Scenario in Large Language Models , author=. 2025 , archivePrefix=
2025
-
[55]
2026 , archivePrefix=
Understanding Users' Privacy Reasoning and Behaviors During Chatbot Use to Support Meaningful Agency in Privacy , author=. 2026 , archivePrefix=
2026
-
[56]
2026 , doi =
Sanitizing manufacturing dataset labels using vision-language models , journal =. 2026 , doi =
2026
-
[58]
QwQ-32B: Embracing the Power of Reinforcement Learning , url =
-
[59]
arXiv preprint arXiv:2412.15115 , year=
Qwen2.5 Technical Report , author=. arXiv preprint arXiv:2412.15115 , year=
-
[60]
Sky-T1: Train your own O1 preview model within \ 450 , howpublished =
-
[61]
Think Less, Achieve More: Cut Reasoning Costs by 50\ howpublished =
-
[62]
Unlocking the Potential of Reinforcement Learning in Improving Reasoning Models , howpublished =
-
[63]
arXiv preprint arXiv:2502.14382 , year=
S*: Test Time Scaling for Code Generation , author=. arXiv preprint arXiv:2502.14382 , year=
-
[64]
arXiv preprint arXiv:2305.11206 , year=
LIMA: Less Is More for Alignment , author=. arXiv preprint arXiv:2305.11206 , year=
-
[66]
arXiv preprint arXiv:2502.11886 , year=
Limr: Less is more for rl scaling , author=. arXiv preprint arXiv:2502.11886 , year=
-
[67]
Bespoke-Stratos: The unreasonable effectiveness of reasoning distillation , howpublished =
-
[68]
Guha, Etash and Raoof, Negin and Mercat, Jean and Frankel, Eric and Keh, Sedrick and Grover, Sachin and Smyrnis, George and Vu, Trung and Marten, Ryan and Saad-Falcon, Jon and Choi, Caroline and Arora, Kushal and Merrill, Mike and Deng, Yichuan and Suvarna, Ashima and Bansal, ...
-
[69]
Open R1: A fully open reproduction of DeepSeek-R1 , url =
-
[70]
arXiv preprint arXiv:2401.10491 , year=
Knowledge fusion of large language models , author=. arXiv preprint arXiv:2401.10491 , year=
-
[71]
arXiv preprint arXiv:2408.07990 , year=
Fusechat: Knowledge fusion of chat models , author=. arXiv preprint arXiv:2408.07990 , year=
-
[72]
arXiv preprint arXiv:2503.04222 , year=
Fusechat-3.0: Preference optimization meets heterogeneous model fusion , author=. arXiv preprint arXiv:2503.04222 , year=
-
[73]
arXiv preprint arXiv:2503.10460 , year=
Light-r1: Curriculum sft, dpo and rl for long cot from scratch and beyond , author=. arXiv preprint arXiv:2503.10460 , year=
-
[74]
arXiv preprint arXiv:2503.04872 , year=
Tinyr1-32b-preview: Boosting accuracy with branch-merge distillation , author=. arXiv preprint arXiv:2503.04872 , year=
-
[76]
The Twelfth International Conference on Learning Representations , year =
Escape Sky-high Cost: Early-stopping Self-Consistency for Multi-step Reasoning , author =. The Twelfth International Conference on Learning Representations , year =
-
[77]
The Fourteenth International Conference on Learning Representations , year =
Deep Think with Confidence , author =. The Fourteenth International Conference on Learning Representations , year =
-
[78]
Advances in Neural Information Processing Systems , volume =
Scalable Best-of- N Selection for Large Language Models via Self-Certainty , author =. Advances in Neural Information Processing Systems , volume =. 2025 , url =
2025
-
[79]
Optimal Aggregation of
Kuang, Peng and Wang, Yanli and Han, Xiaoyu and Liu, Yaowenqi and Xu, Kaidi and Wang, Haohan , booktitle =. Optimal Aggregation of. 2026 , url =
2026
-
[80]
Advances in Neural Information Processing Systems , volume =
Majority of the Bests: Improving Best-of- N via Bootstrapping , author =. Advances in Neural Information Processing Systems , volume =. 2025 , url =
2025
-
[81]
The Fourteenth International Conference on Learning Representations , year =
Best-of-Majority: Minimax-Optimal Strategy for Pass@ k Inference Scaling , author =. The Fourteenth International Conference on Learning Representations , year =
-
[82]
Transactions on Machine Learning Research , year =
From Decoding to Meta-Generation: Inference-time Algorithms for Large Language Models , author =. Transactions on Machine Learning Research , year =
-
[83]
Proceedings of the 40th International Conference on Machine Learning (ICML) , series =
Scaling Laws for Reward Model Overoptimization , author =. Proceedings of the 40th International Conference on Machine Learning (ICML) , series =. 2023 , url =
2023
-
[84]
Proceedings of the 42nd International Conference on Machine Learning (ICML) , series =
Is Best-of- N the Best of Them? Coverage, Scaling, and Optimality in Inference-Time Alignment , author =. Proceedings of the 42nd International Conference on Machine Learning (ICML) , series =. 2025 , url =
2025
- [86]
- [87]
-
[88]
Proceedings of the Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics (HLT-NAACL) , year =
Minimum Bayes-Risk Decoding for Statistical Machine Translation , author =. Proceedings of the Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics (HLT-NAACL) , year =
-
[89]
Transactions of the Association for Computational Linguistics , volume =
High Quality Rather than High Model Probability: Minimum Bayes Risk Decoding with Neural Metrics , author =. Transactions of the Association for Computational Linguistics , volume =. 2022 , doi =
2022
- [90]
- [91]
- [92]
-
[93]
Eikema, Bryan and Aziz, Wilker , booktitle =. Is. 2020 , url =
2020
- [94]
- [95]
- [96]
- [97]
- [98]
- [99]
- [100]
- [102]
-
[103]
2024 , month = jul, publisher =
Wan, Ziyu and Feng, Xidong and Wen, Muning and Mcaleer, Stephen Marcus and Wen, Ying and Zhang, Weinan and Wang, Jun , booktitle =. 2024 , month = jul, publisher =
2024
-
[104]
2024 , month = jun, publisher =
Yu, Fei and Gao, Anningzhe and Wang, Benyou , booktitle =. 2024 , month = jun, publisher =. doi:10.18653/v1/2024.findings-naacl.55 , url =
2024 doi
-
[105]
Policy Guided Tree Search for Enhanced
Li, Yang , booktitle =. Policy Guided Tree Search for Enhanced. 2025 , month = jul, publisher =
2025
-
[106]
Transactions on Machine Learning Research , year =
Tree Search for Language Model Agents , author =. Transactions on Machine Learning Research , year =
-
[107]
2026 , eprint =
Policy-Guided Search on Tree-of-Thoughts for Efficient Problem Solving with Bounded Language Model Queries , author =. 2026 , eprint =. doi:10.48550/arXiv.2601.03606 , url =
2026 doi
-
[108]
2026 , note =
On the Scaling Flaws of Verifier-Guided Beam Search , author =. 2026 , note =
2026
-
[109]
Advances in Neural Information Processing Systems , year =
Reasoning Is Not a Race: When Stopping Early Beats Going Deeper , author =. Advances in Neural Information Processing Systems , year =
-
[110]
DEFT: Decoding with Flash Tree-Attention for Efficient Tree-Structured
Yao, Jinwei and Chen, Kaiqi and Zhang, Kexun and You, Jiaxuan and Yuan, Binhang and Wang, Zeke and Lin, Tao , booktitle =. DEFT: Decoding with Flash Tree-Attention for Efficient Tree-Structured. 2025 , url =
2025
-
[111]
Advances in Neural Information Processing Systems , year =
Thought of Search: Planning with Language Models Through The Lens of Efficiency , author =. Advances in Neural Information Processing Systems , year =
-
[112]
Bandit Based
Kocsis, Levente and Szepesv. Bandit Based. Proceedings of the 17th European Conference on Machine Learning (ECML) , year =
-
[113]
and Powley, Edward and Whitehouse, Daniel and Lucas, Simon M
Browne, Cameron B. and Powley, Edward and Whitehouse, Daniel and Lucas, Simon M. and Cowling, Peter I. and Rohlfshagen, Philipp and Tavener, Stephen and Perez, Diego and Samothrakis, Spyridon and Colton, Simon , journal =. A Survey of. 2012 , doi =
2012
-
[114]
Proceedings of the 2025 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages =
Reward-Guided Tree Search for Inference Time Alignment of Large Language Models , author =. Proceedings of the 2025 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages =. 2025 , url =
2025
-
[115]
2024 , urldate =
American Invitational Mathematics Examination (AIME) , author =. 2024 , urldate =
2024
-
[116]
2025 , urldate =
American Invitational Mathematics Examination (AIME) , author =. 2025 , urldate =
2025
-
[117]
2025 , urldate =
HMMT February 2025 Archive (Problems and Solutions) , author =. 2025 , urldate =
2025
-
[118]
2025 , urldate =
Brown University Math Olympiad (BrUMO) , author =. 2025 , urldate =
2025
-
[119]
NovaSky Team , title =
-
[120]
2025 , eprint=
Qwen3 Technical Report , author=. 2025 , eprint=
2025
-
[121]
2025 , eprint=
gpt-oss-120b & gpt-oss-20b Model Card , author=. 2025 , eprint=
2025
-
[122]
2502.03387 , archivePrefix=
Yixin Ye and Zhen Huang and Yang Xiao and Ethan Chern and Shijie Xia and Pengfei Liu , year=. 2502.03387 , archivePrefix=
-
[123]
arXiv preprint arXiv:2507.11407 , year=
EXAONE 4.0: Unified Large Language Models Integrating Non-reasoning and Reasoning Modes , author=. arXiv preprint arXiv:2507.11407 , year=
-
[124]
2025 , eprint=
Wasi Uddin Ahmad and Sean Narenthiran and Somshubra Majumdar and Aleksander Ficek and Siddhartha Jain and Jocelyn Huang and Vahid Noroozi and Boris Ginsburg , journal=. 2025 , eprint=
2025
-
[126]
2504.16891 , archivePrefix=
Ivan Moshkov and Darragh Hanley and Ivan Sorokin and Shubham Toshniwal and Christof Henkel and Benedikt Schifferer and Wei Du and Igor Gitman , year=. 2504.16891 , archivePrefix=
-
[127]
2025 , url=
Shubham Toshniwal and Ivan Sorokin and Aleksander Ficek and Ivan Moshkov and Igor Gitman , booktitle=. 2025 , url=
2025
-
[128]
and Zhang, Songyang and Chen, Kai , title =
Liu, Shudong and Liu, Hongwei and Liu, Junnan and Xiao, Linchen and Gao, Songyang and Lyu, Chengqi and Gu, Yuzhe and Zhang, Wenwei and Wong, Derek F. and Zhang, Songyang and Chen, Kai , title =. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Proces...
2025
-
[130]
Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=
Fusechat: Knowledge fusion of chat models , author=. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=
2025
-
[131]
Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track) , pages=
Light-r1: Curriculum sft, dpo and rl for long cot from scratch and beyond , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track) , pages=
-
[132]
arXiv preprint arXiv:2506.13284 , year=
AceReason-Nemotron 1.1: Advancing Math and Code Reasoning through SFT and RL Synergy , author=. arXiv preprint arXiv:2506.13284 , year=
-
[134]
Advances in Neural Information Processing Systems , year =
70\ author =. Advances in Neural Information Processing Systems , year =
-
[135]
arXiv preprint arXiv:2307.08691 , year=
Flashattention-2: Faster attention with better parallelism and work partitioning , author=. arXiv preprint arXiv:2307.08691 , year=
-
[136]
, author =
Accelerate: Training and inference at scale made simple, efficient and adaptable. , author =
-
[137]
arXiv preprint arXiv:1910.03771 , year=
Huggingface's transformers: State-of-the-art natural language processing , author=. arXiv preprint arXiv:1910.03771 , year=
1910 arXiv
-
[138]
and Zhang, Hao and Stoica, Ion , title =
Kwon, Woosuk and Li, Zhuohan and Zhuang, Siyuan and Sheng, Ying and Zheng, Lianmin and Yu, Cody Hao and Gonzalez, Joseph E. and Zhang, Hao and Stoica, Ion , title =. Proceedings of the 29th Symposium on Operating Systems Principles , pages =. 2023 , publisher =. doi:10.1145/36...
2023
-
[139]
Proceedings of the 14th International Conference on Learning Representations (ICLR 2026) , year =
Don't Pass@k: A Bayesian Framework for Large Language Model Evaluation , author =. Proceedings of the 14th International Conference on Learning Representations (ICLR 2026) , year =. 2510.04265 , archivePrefix =
2026 arXiv
-
[140]
Competition-level code generation with
Li, Yujia and Choi, David and others , journal=. Competition-level code generation with. 2022 , doi=
2022
-
[141]
arXiv preprint arXiv:2408.00724 , year=
Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models , author=. arXiv preprint arXiv:2408.00724 , year=
-
[142]
Snell, Charlie and Lee, Jaehoon and Xu, Kelvin and Kumar, Aviral , booktitle=. Scaling
-
[143]
Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year=
Improving Minimum Bayes Risk Decoding with Multi-Prompt , author=. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP) , year=
2024
-
[144]
arXiv preprint arXiv:2310.02743 , year=
Reward Model Ensembles Help Mitigate Overoptimization , author=. arXiv preprint arXiv:2310.02743 , year=
-
[145]
Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL) , pages=
Revisiting the Test-Time Scaling of o1-like Models: Do they Truly Possess Test-Time Scaling Capabilities? , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL) , pages=
-
[146]
arXiv preprint arXiv:2512.02008 , year=
The Art of Scaling Test-Time Compute for Large Language Models , author=. arXiv preprint arXiv:2512.02008 , year=
-
[147]
Confidence Improves Self-Consistency in
Taubenfeld, Amir and Sheffer, Tom and Ofek, Eran and Feder, Amir and Goldstein, Ariel and Gekhman, Zorik and Yona, Gal , booktitle =. Confidence Improves Self-Consistency in. 2025 , month = jul, publisher =
2025
-
[148]
2024 , eprint =
Semantic Self-Consistency: Enhancing Language Model Reasoning via Semantic Weighting , author =. 2024 , eprint =
2024
-
[149]
and Zhang, Hao and Gonzalez, Joseph E
Zheng, Lianmin and Chiang, Wei-Lin and Sheng, Ying and Zhuang, Siyuan and Wu, Zhanghao and Zhuang, Yonghao and Lin, Zi and Li, Zhuohan and Li, Dacheng and Xing, Eric P. and Zhang, Hao and Gonzalez, Joseph E. and Stoica, Ion , year =. Judging. 2306.05685 , archivePrefix=
-
[150]
2023 , month = dec, address =
Liu, Yang and Iter, Dan and Xu, Yichong and Wang, Shuohang and Xu, Ruochen and Zhu, Chenguang , booktitle =. 2023 , month = dec, address =
2023
-
[151]
Evaluating Judges as Evaluators: The
Zhou, Yilun and Xu, Austin and Wang, Peifeng and Xiong, Caiming and Joty, Shafiq , year =. Evaluating Judges as Evaluators: The. 2504.15253 , archivePrefix=
-
[152]
PairJudge
Liu, Yantao and Yao, Zijun and Min, Rui and Cao, Yixin and Hou, Lei and Li, Juanzi , year =. PairJudge. 2501.13007 , archivePrefix=
-
[153]
2025 , eprint =
Inference-Time Reward Hacking in Large Language Models , author =. 2025 , eprint =
2025
-
[154]
2025 , eprint =
Inference-Time Scaling for Generalist Reward Modeling , author =. 2025 , eprint =
2025
-
[155]
Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages =
Answering Questions by Meta-Reasoning over Multiple Chains of Thought , author =. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages =. 2023 , address =
2023
- [156]
- [157]
-
[158]
Advances in Neural Information Processing Systems , year =
Training Language Models to Follow Instructions with Human Feedback , author =. Advances in Neural Information Processing Systems , year =
- [159]
-
[161]
2026 , archivePrefix=
REALM: Reliable Expertise-Aware Language Model Fine-Tuning from Noisy Annotations , author=. 2026 , archivePrefix=
2026
- [162]
-
[163]
2025 , eprint =
EXAONE 4.0: Unified Large Language Models Integrating Non-reasoning and Reasoning Modes , author =. 2025 , eprint =. doi:10.48550/arXiv.2507.11407 , url =
2025 doi
-
[164]
deepseek-ai/DeepSeek-R1-Distill-Qwen-32B (Model Card) , howpublished =
-
[165]
DeepSeek-R1 (GitHub Repository) , howpublished =
-
[166]
Qwen/QwQ-32B (Model Card) , howpublished =
- [167]
- [168]
-
[169]
Qwen/Qwen3-4B-Thinking-2507 (Model Card) , howpublished =
-
[170]
Qwen/Qwen3-30B-A3B-Thinking-2507 (Model Card) , howpublished =
-
[171]
Qwen/Qwen3-Next-80B-A3B-Thinking (Model Card) , howpublished =
-
[172]
gpt-oss-120b & gpt-oss-20b Model Card , howpublished =
- [173]
-
[174]
Introducing gpt-oss , howpublished =
-
[175]
openai/gpt-oss-20b (Model Card) , howpublished =
-
[176]
gpt-oss-20b Model (OpenAI API Documentation) , howpublished =
-
[177]
OpenAI open-weight models (gpt-oss) , howpublished =
-
[178]
openai/gpt-oss-safeguard-20b (Model Card) , howpublished =
- [179]
- [180]
- [181]
- [182]
-
[183]
Bespoke-Stratos-17k Dataset Card , howpublished =
-
[184]
Bespoke-Stratos-32B Model Card , howpublished =
-
[185]
Bespoke-Stratos-7B Model Card , howpublished =
-
[186]
Bespoke Curator: Synthetic Data Curation for Post-Training , howpublished =
- [187]
-
[188]
OpenThoughts-114K Dataset Card , howpublished =
-
[189]
OpenThoughts2-1M Dataset Card , howpublished =
-
[190]
OpenThoughts3-1.2M Dataset Card , howpublished =
-
[191]
OpenThinker2-32B Model Card , howpublished =
-
[192]
OpenThinker3-7B Model Card , howpublished =
- [193]
- [194]
-
[195]
microsoft/Phi-4-mini-flash-reasoning (Model Card) , howpublished =
-
[196]
2025 , eprint =
Decoder-Hybrid-Decoder Architecture for Efficient Reasoning with Long Generation , author =. 2025 , eprint =. doi:10.48550/arXiv.2507.06607 , url =
2025 doi
-
[197]
Open-R1: A fully open reproduction of DeepSeek-R1 , author =
-
[198]
Mixture-of-Thoughts Dataset Card , author =
-
[199]
OpenR1-Distill-7B Model Card , author =
-
[200]
OpenR1-Math-220k Dataset Card , author =
-
[201]
Open R1: Update \#2 , author =
-
[202]
CodeForces-CoTs Dataset Card , author =
-
[203]
OlympicCoder-7B Model Card , author =
-
[204]
OlympicCoder-32B Model Card , author =
-
[205]
Open R1: Update \#3 , author =
-
[206]
LIMO: Less is More for Reasoning (code and resources) , howpublished =
-
[207]
LIMR: Less is More for RL Scaling (code and resources) , howpublished =
-
[208]
GAIR/LIMO-v2 (Model Card) , howpublished =
-
[209]
GAIR/LIMO-v2 (Dataset Card) , howpublished =
-
[210]
LIMOPro: Reasoning Refinement for Efficient and Effective Test-time Scaling , howpublished =
-
[211]
FuseAI: Model Fusion for Large Language Models (code and resources) , howpublished =
-
[212]
FuseO1-DeepSeekR1-QwQ-SkyT1-32B-Preview (Model Card) , howpublished =
-
[213]
FuseO1-DeepSeekR1-QwQ-SkyT1-Flash-32B-Preview (Model Card) , howpublished =
-
[214]
mergekit: Tools for merging pretrained large language models , howpublished =
-
[215]
qihoo360/Light-R1-32B (Model Card) , howpublished =
-
[216]
qihoo360/Light-R1-7B-DS (Model Card) , howpublished =
-
[217]
qihoo360/Light-R1-32B-DS (Model Card) , howpublished =
-
[218]
qihoo360/Light-R1-14B-DS (Model Card) , howpublished =
-
[219]
Journal of the American Statistical Association , volume =
Probable Inference, the Law of Succession, and Statistical Inference , author =. Journal of the American Statistical Association , volume =. 1927 , doi =
1927
-
[220]
Are Your
Junnan Liu and Hongwei Liu and Linchen Xiao and Ziyi Wang and Kuikun Liu and Songyang Gao and Wenwei Zhang and Songyang Zhang and Kai Chen , booktitle =. Are Your. 2025 , pages =. doi:10.18653/v1/2025.findings-acl.905 , url =
2025 doi
-
[221]
International Conference on Learning Representations , year=
LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code , author=. International Conference on Learning Representations , year=
-
[222]
Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=
Hierarchical Neural Story Generation , author=. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=
-
[223]
International Conference on Learning Representations , year=
The Curious Case of Neural Text Degeneration , author=. International Conference on Learning Representations , year=
-
[224]
Transactions of the Association for Computational Linguistics , volume=
Locally Typical Sampling , author=. Transactions of the Association for Computational Linguistics , volume=
-
[225]
2025 , month = mar, howpublished =
Sutawika, Lintang and Schoelkopf, Hailey and Gao, Leo and Abbasi, Baber and Biderman, Stella and Tow, Jonathan and others , title =. 2025 , month = mar, howpublished =. doi:10.5281/zenodo.14970487 , url =
2025 doi
-
[226]
Advances in Neural Information Processing Systems , volume =
Balunovic, Mislav and Dekoninck, Jasper and Petrov, Ivo and Jovanovi. Advances in Neural Information Processing Systems , volume =. 2025 , url =
2025
- [227]
-
[228]
2025 , url =
Advances in Neural Information Processing Systems , volume =. 2025 , url =
2025
-
[229]
2025 , howpublished =
2025
- [230]
-
[231]
2026 , month = jun, howpublished =
2026
-
[232]
2024 , howpublished =
Jia, Maxwell , title =. 2024 , howpublished =
2024
- [233]
-
[234]
2026 , howpublished =
2026
-
[235]
Phi-4-reasoning technical report
Marah Abdin, Sahaj Agarwal, Ahmed Awadallah, Vidhisha Balachandran, Harkirat Behl, Lingjiao Chen, Gustavo de Rosa, Suriya Gunasekar, Mojan Javaheripi, Neel Joshi, et al. Phi-4-reasoning technical report. arXiv preprint arXiv:2504.21318, 2025. doi:10.48550/arXiv.2504.21318. URL...
-
[236]
Phi-4 technical report, 2024
Marah Abdin et al. Phi-4 technical report, 2024. URL https://arxiv.org/abs/2412.08905
2024 arXiv
-
[237]
Let ' s sample step by step: Adaptive-consistency for efficient reasoning and coding with LLM s
Pranjal Aggarwal, Aman Madaan, Yiming Yang, and Mausam. Let ' s sample step by step: Adaptive-consistency for efficient reasoning and coding with LLM s. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pp.\ 12375--12396, Singapore, Dec...
2023 doi
-
[238]
OpenCodeReasoning-II: A Simple Test Time Scaling Approach via Self-Critique , 2025 a
Wasi Uddin Ahmad, Somshubra Majumdar, Aleksander Ficek, Sean Narenthiran, Mehrzad Samadi, Jocelyn Huang, Siddhartha Jain, Vahid Noroozi, and Boris Ginsburg. OpenCodeReasoning-II: A Simple Test Time Scaling Approach via Self-Critique , 2025 a . URL https://arxiv.org/abs/2507.09075
2025 arXiv
-
[239]
OpenCodeReasoning: Advancing Data Distillation for Competitive Coding
Wasi Uddin Ahmad, Sean Narenthiran, Somshubra Majumdar, Aleksander Ficek, Siddhartha Jain, Jocelyn Huang, Vahid Noroozi, and Boris Ginsburg. OpenCodeReasoning: Advancing Data Distillation for Competitive Coding . arXiv preprint arXiv:2504.01943, 2025 b . URL https://arxiv.org/...
2025 arXiv
-
[240]
Back to basics: Revisiting reinforce style optimization for learning from human feedback in llms, 2024
Arash Ahmadian, Chris Cremer, Matthias Gall \'e , Marzieh Fadaee, Julia Kreutzer, Olivier Pietquin, Ahmet \"U st \"u n, and Sara Hooker. Back to basics: Revisiting reinforce style optimization for learning from human feedback in llms, 2024. URL https://arxiv.org/abs/2402.14740
2024 arXiv
-
[241]
mergekit: Tools for merging pretrained large language models
Arcee AI . mergekit: Tools for merging pretrained large language models. https://github.com/arcee-ai/mergekit, 2024. Accessed: 2026-02-26
2024
-
[242]
MathArena : Evaluating llms on uncontaminated math competitions
Mislav Balunovic, Jasper Dekoninck, Ivo Petrov, Nikola Jovanovi \'c , and Martin Vechev. MathArena : Evaluating llms on uncontaminated math competitions. In Advances in Neural Information Processing Systems, volume 38, 2025. URL https://proceedings.neurips.cc/paper_files/paper...
2025
-
[243]
Nvidia nemotron nano 2: An accurate and efficient hybrid mamba-transformer reasoning model
Aarti Basant, Abhijit Khairnar, Abhijit Paithankar, Abhinav Khattar, Adithya Renduchintala, Aditya Malte, Akhiad Bercovich, Akshay Hazare, Alejandra Rico, Aleksander Ficek, et al. Nvidia nemotron nano 2: An accurate and efficient hybrid mamba-transformer reasoning model. arXiv...
2025 arXiv
-
[244]
Amanda Bertsch, Alex Xie, Graham Neubig, and Matthew R. Gormley. It's MBR all the way down: Modern generation techniques through the lens of minimum bayes risk, 2023. URL https://arxiv.org/abs/2310.01387
2023 arXiv
-
[245]
Bespoke curator: Synthetic data curation for post-training
Bespoke Labs . Bespoke curator: Synthetic data curation for post-training. https://github.com/bespokelabsai/curator, 2025 a . Accessed: 2025-01-31
2025
-
[246]
Bespoke-stratos: The unreasonable effectiveness of reasoning distillation
Bespoke Labs . Bespoke-stratos: The unreasonable effectiveness of reasoning distillation. https://www.bespokelabs.ai/blog/bespoke-stratos-the-unreasonable-effectiveness-of-reasoning-distillation, 2025 b . Accessed: 2025-01-22
2025
-
[247]
Bespoke-stratos-17k dataset card
Bespoke Labs . Bespoke-stratos-17k dataset card. https://huggingface.co/datasets/bespokelabs/Bespoke-Stratos-17k, 2025 c . Accessed: 2025-01-31
2025
-
[248]
Bespoke-stratos-32b model card
Bespoke Labs . Bespoke-stratos-32b model card. https://huggingface.co/bespokelabs/Bespoke-Stratos-32B, 2025 d . Accessed: 2025-01-31
2025
-
[249]
Bespoke-stratos-7b model card
Bespoke Labs . Bespoke-stratos-7b model card. https://huggingface.co/bespokelabs/Bespoke-Stratos-7B, 2025 e . Accessed: 2025-01-31
2025
- [250]
-
[251]
Towards reproducible llm evaluation: Quantifying uncertainty in llm benchmark scores
Robert E Blackwell, Jon Barry, and Anthony G Cohn. Towards reproducible llm evaluation: Quantifying uncertainty in llm benchmark scores. arXiv preprint arXiv:2410.03492, 2024
2024 arXiv
-
[252]
Accounting for variance in machine learning benchmarks
Xavier Bouthillier, Pierre Delaunay, Mirko Bronzi, Assya Trofimov, Brennan Nichyporuk, Justin Szeto, Nazanin Mohammadi Sepahvand, Edward Raff, Kanika Madan, Vikram Voleti, Samira Ebrahimi Kahou, Vincent Michalski, Dmitriy Serdyuk, Tal Arbel, Chris Pal, Ga \"e l Varoquaux, and ...
2021
-
[253]
Browne, Edward Powley, Daniel Whitehouse, Simon M
Cameron B. Browne, Edward Powley, Daniel Whitehouse, Simon M. Lucas, Peter I. Cowling, Philipp Rohlfshagen, Stephen Tavener, Diego Perez, Spyridon Samothrakis, and Simon Colton. A survey of M onte C arlo tree search methods. IEEE Transactions on Computational Intelligence and ...
2012
-
[254]
Evaluating large language models trained on code
Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde De Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, et al. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374, 2021
2021 arXiv
-
[255]
Universal self-consistency for large language model generation, 2023
Xinyun Chen, Renat Aksitov, Uri Alon, Jie Ren, Kefan Xiao, Pengcheng Yin, Sushant Prakash, Charles Sutton, Xuezhi Wang, and Denny Zhou. Universal self-consistency for large language model generation, 2023. URL https://arxiv.org/abs/2311.17311
2023 arXiv
-
[256]
Faster minimum bayes risk decoding with confidence-based pruning, 2023
Julius Cheng and Andreas Vlachos. Faster minimum bayes risk decoding with confidence-based pruning, 2023. URL https://arxiv.org/abs/2311.14919
2023 arXiv
-
[257]
Chatbot arena: An open platform for evaluating llms by human preference
Wei-Lin Chiang, Lianmin Zheng, Ying Sheng, Anastasios Nikolas Angelopoulos, Tianle Li, Dacheng Li, Banghua Zhu, Hao Zhang, Michael Jordan, Joseph E Gonzalez, et al. Chatbot arena: An open platform for evaluating llms by human preference. In Forty-first International Conference...
2024
-
[258]
Training verifiers to solve math word problems, 2021
Karl Cobbe et al. Training verifiers to solve math word problems, 2021. URL https://arxiv.org/abs/2110.14168
2021 arXiv
-
[259]
Process reinforcement through implicit rewards, 2025
Ganqu Cui, Lifan Yuan, Zefan Wang, Hanbin Wang, Yuchen Zhang, Jiacheng Chen, Wendi Li, Bingxiang He, Yuchen Fan, Tianyu Yu, Qixin Xu, Weize Chen, Jiarui Yuan, Huayu Yuan, Kaiyan Zhang, Xingtai Lv, Shuo Wang, Yuan Yao, Xu Han, Hao Peng, Yu Cheng, Zhiyuan Liu, Maosong Sun, Bowen...
2025 arXiv
-
[260]
deepseek-ai/deepseek-r1-distill-qwen-32b (model card)
DeepSeek-AI . deepseek-ai/deepseek-r1-distill-qwen-32b (model card). https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B, 2025 a . Accessed: 2026-02-26
2025
-
[261]
Deepseek-r1 (github repository)
DeepSeek-AI . Deepseek-r1 (github repository). https://github.com/deepseek-ai/DeepSeek-R1, 2025 b . Accessed: 2026-02-26
2025
-
[262]
Beyond benchmarks: MathArena as an evaluation platform for mathematics with llms, 2026
Jasper Dekoninck, Nikola Jovanovi \'c , Tim Gehrunger, K \'a ri R \"o gnvaldsson, Ivo Petrov, Chenhao Sun, and Martin Vechev. Beyond benchmarks: MathArena as an evaluation platform for mathematics with llms, 2026. URL https://arxiv.org/abs/2605.00674
2026 arXiv
-
[263]
Best-of-majority: Minimax-optimal strategy for pass@ k inference scaling
Qiwei Di, Kaixuan Ji, Xuheng Li, Heyang Zhao, and Quanquan Gu. Best-of-majority: Minimax-optimal strategy for pass@ k inference scaling. In The Fourteenth International Conference on Learning Representations, 2026. URL https://openreview.net/forum?id=a6CVQpjbXq
2026
-
[264]
Jesse Dodge, Suchin Gururangan, Dallas Card, Roy Schwartz, and Noah A. Smith. Show your work: Improved reporting of experimental results. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Nat...
2019 doi
-
[265]
Bootstrap methods: Another look at the jackknife
Bradley Efron. Bootstrap methods: Another look at the jackknife. The Annals of Statistics, 7 0 (1): 0 1--26, 1979
1979
-
[266]
Is MAP decoding all you need? the inadequacy of the mode in neural machine translation
Bryan Eikema and Wilker Aziz. Is MAP decoding all you need? the inadequacy of the mode in neural machine translation. In Proceedings of the 28th International Conference on Computational Linguistics (COLING), 2020. URL https://aclanthology.org/2020.coling-main.398/
2020
-
[267]
Hierarchical neural story generation
Angela Fan, Mike Lewis, and Yann Dauphin. Hierarchical neural story generation. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp.\ 889--898, 2018
2018
-
[268]
High quality rather than high model probability: Minimum bayes risk decoding with neural metrics
Markus Freitag, David Grangier, Qijun Tan, and Bowen Liang. High quality rather than high model probability: Minimum bayes risk decoding with neural metrics. Transactions of the Association for Computational Linguistics, 10: 0 811--825, 2022. doi:10.1162/tacl_a_00491. URL http...
2022 doi
-
[269]
Deep think with confidence
Yichao Fu, Xuewei Wang, Hao Zhang, Yuandong Tian, and Jiawei Zhao. Deep think with confidence. In The Fourteenth International Conference on Learning Representations, 2026. URL https://openreview.net/forum?id=8LqHs0KIM7
2026
-
[270]
Fuseo1-deepseekr1-qwq-skyt1-32b-preview (model card)
FuseAI . Fuseo1-deepseekr1-qwq-skyt1-32b-preview (model card). https://huggingface.co/FuseAI/FuseO1-DeepSeekR1-QwQ-SkyT1-32B-Preview, 2025 a . Accessed: 2026-02-26
2025
-
[271]
Fuseo1-deepseekr1-qwq-skyt1-flash-32b-preview (model card)
FuseAI . Fuseo1-deepseekr1-qwq-skyt1-flash-32b-preview (model card). https://huggingface.co/FuseAI/FuseO1-DeepSeekR1-QwQ-SkyT1-Flash-32B-Preview, 2025 b . Accessed: 2026-02-26
2025
-
[272]
Fuseai: Model fusion for large language models (code and resources)
FuseAI . Fuseai: Model fusion for large language models (code and resources). https://github.com/fanqiwan/FuseAI, 2025 c . Accessed: 2026-02-26
2025
-
[273]
Gair/limo-v2 (dataset card)
GAIR . Gair/limo-v2 (dataset card). https://huggingface.co/datasets/GAIR/LIMO-v2, 2025 a . Accessed: 2026-02-26
2025
-
[274]
Gair/limo-v2 (model card)
GAIR . Gair/limo-v2 (model card). https://huggingface.co/GAIR/LIMO-v2, 2025 b . Accessed: 2026-02-26
2025
-
[275]
Scaling laws for reward model overoptimization
Leo Gao, John Schulman, and Jacob Hilton. Scaling laws for reward model overoptimization. In Proceedings of the 40th International Conference on Machine Learning (ICML), volume 202 of Proceedings of Machine Learning Research, pp.\ 10835--10866, 2023. URL https://proceedings.ml...
2023
-
[276]
Decentralized low-rank fine-tuning of large language models
Sajjad Ghiasvand, Mahnoosh Alizadeh, and Ramtin Pedarsani. Decentralized low-rank fine-tuning of large language models. In Proceedings of the 1st Workshop for Research on Agent Language Models (REALM 2025). Association for Computational Linguistics, July 2025. doi:10.18653/v1/...
2025 doi
-
[277]
Realm: Reliable expertise-aware language model fine-tuning from noisy annotations, 2026
Sajjad Ghiasvand, Mark Beliaev, Mahnoosh Alizadeh, and Ramtin Pedarsani. Realm: Reliable expertise-aware language model fine-tuning from noisy annotations, 2026. URL https://arxiv.org/abs/2604.17289
2026 arXiv
-
[278]
Evalchemy , November 2024
Etash Guha, Negin Raoof, Jean Mercat, Eric Frankel, Sedrick Keh, Sachin Grover, George Smyrnis, Trung Vu, Ryan Marten, Jon Saad-Falcon, Caroline Choi, Kushal Arora, Mike Merrill, Yichuan Deng, Ashima Suvarna, Hritik Bansal, Marianna Nezhurina, Yejin Choi, Reinhard Heckel, Seew...
2024
-
[279]
Openthoughts: Data recipes for reasoning models, 2025
Etash Guha, Ryan Marten, Sedrick Keh, Negin Raoof, Georgios Smyrnis, Hritik Bansal, Marianna Nezhurina, Jean Mercat, Trung Vu, Zayne Sprague, et al. Openthoughts: Data recipes for reasoning models, 2025. URL https://arxiv.org/abs/2506.04178
2025 arXiv
-
[280]
Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning
Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948, 2025
2025 arXiv
-
[281]
Reasoning with language model is planning with world model
Shibo Hao, Yi Gu, Haodi Ma, Joshua Hong, Zhen Wang, Daisy Zhe Wang, and Zhiting Hu. Reasoning with language model is planning with world model. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pp.\ 8154--8173, Singapore, December 2023....
2023 doi
-
[282]
Ranking reasoning LLM s under test-time scaling
Mohsen Hariri, Michael Hinczewski, Jing Ma, and Vipin Chaudhary. Ranking reasoning LLM s under test-time scaling. In Maria Liakata, Viviane P. Moreira, Jiajun Zhang, and David Jurgens (eds.), Proceedings of the 64th Annual Meeting of the A ssociation for C omputational L ingui...
2026 doi
-
[283]
Quantize what counts: More for keys, less for values
Mohsen Hariri, Alan Luo, Weicong Chen, Tianyi Zhang, Qifan Wang, Xiaotian Han, and Vipin Chaudhary. Quantize what counts: More for keys, less for values. In Maria Liakata, Viviane P. Moreira, Jiajun Zhang, and David Jurgens (eds.), Findings of the A ssociation for C omputation...
2026 doi
-
[284]
Don't pass@k: A bayesian framework for large language model evaluation
Mohsen Hariri, Amirhossein Samandar, Michael Hinczewski, and Vipin Chaudhary. Don't pass@k: A bayesian framework for large language model evaluation. In Proceedings of the 14th International Conference on Learning Representations (ICLR 2026), 2026 c . URL https://openreview.ne...
2026
-
[285]
Truncation sampling as language model desmoothing
John Hewitt, Christopher D Manning, and Percy Liang. Truncation sampling as language model desmoothing. arXiv preprint arXiv:2210.15191, 2022
2022 arXiv
-
[286]
A sober look at progress in language model reasoning: Pitfalls and paths to reproducibility
Andreas Hochlehnert, Hardik Bhatnagar, Vishaal Udandarao, Samuel Albanie, Ameya Prabhu, and Matthias Bethge. A sober look at progress in language model reasoning: Pitfalls and paths to reproducibility. In Second Conference on Language Modeling, 2025. URL https://openreview.net...
2025
-
[287]
The curious case of neural text degeneration
Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. The curious case of neural text degeneration. In International Conference on Learning Representations, 2020
2020
-
[288]
Is best-of- n the best of them? coverage, scaling, and optimality in inference-time alignment
Audrey Huang, Adam Block, Qinghua Liu, Nan Jiang, Akshay Krishnamurthy, and Dylan Foster. Is best-of- n the best of them? coverage, scaling, and optimality in inference-time alignment. In Proceedings of the 42nd International Conference on Machine Learning (ICML), volume 267 o...
-
[289]
Open r1: A fully open reproduction of deepseek-r1, January 2025 a
Hugging Face . Open r1: A fully open reproduction of deepseek-r1, January 2025 a . URL https://github.com/huggingface/open-r1
2025
-
[290]
Open-r1: A fully open reproduction of deepseek-r1
Hugging Face . Open-r1: A fully open reproduction of deepseek-r1. https://github.com/huggingface/open-r1, 2025 b . Accessed: 2026-02-26
2025
-
[291]
Open r1: Update \#2
Hugging Face . Open r1: Update \#2. https://huggingface.co/blog/open-r1/update-2, 2025 c . Accessed: 2026-02-26
2025
-
[292]
Open r1: Update \#3
Hugging Face . Open r1: Update \#3. https://huggingface.co/blog/open-r1/update-3, 2025 d . Accessed: 2026-02-26
2025
-
[293]
Reward-guided tree search for inference time alignment of large language models
Chia-Yu Hung, Navonil Majumder, Ambuj Mehrish, and Soujanya Poria. Reward-guided tree search for inference time alignment of large language models. In Proceedings of the 2025 Conference of the North American Chapter of the Association for Computational Linguistics: Human Langu...
2025
-
[294]
Wang, Armando Solar-Lezama, Koushik Sen, and Ion Stoica
Naman Jain, King Han, Alex Gu, Wen-Ding Li, Fanjia Yan, Tianjun Zhang, Sida I. Wang, Armando Solar-Lezama, Koushik Sen, and Ion Stoica. Livecodebench: Holistic and contamination free evaluation of large language models for code. In International Conference on Learning Represen...
2025
-
[295]
AIME 2024 dataset
Maxwell Jia. AIME 2024 dataset. Hugging Face dataset, 2024. URL https://huggingface.co/datasets/Maxwell-Jia/AIME_2024
2024
-
[296]
Hyperparameter-free approach for faster minimum bayes risk decoding, 2024
Yuu Jinnai and Kaito Ariu. Hyperparameter-free approach for faster minimum bayes risk decoding, 2024. URL https://arxiv.org/abs/2401.02749
2024 arXiv
-
[297]
Regularized best-of- n sampling with minimum B ayes risk objective for language model alignment
Yuu Jinnai, Tetsuro Morimura, Kaito Ariu, and Kenshi Abe. Regularized best-of- n sampling with minimum B ayes risk objective for language model alignment. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguist...
2025 doi
-
[298]
Scalable best-of- n selection for large language models via self-certainty
Zhewei Kang, Xuandong Zhao, and Dawn Song. Scalable best-of- n selection for large language models via self-certainty. In Advances in Neural Information Processing Systems, volume 38, 2025. URL https://proceedings.neurips.cc/paper_files/paper/2025/hash/1c7eff166a8e345f664f0faa...
2025
-
[299]
Inference-time reward hacking in large language models, 2025
Hadi Khalaf, Claudio Mayrink Verdun, Alex Oesterling, Himabindu Lakkaraju, and Flavio du Pin Calmon. Inference-time reward hacking in large language models, 2025. URL https://arxiv.org/abs/2506.19248. Accepted to NeurIPS 2025 (Spotlight)
2025
-
[300]
Simulating a bias mitigation scenario in large language models, 2025
Kiana Kiashemshaki, Mohammad Jalili Torkamani, Negin Mahmoudi, and Meysam Shirdel Bilehsavar. Simulating a bias mitigation scenario in large language models, 2025. URL https://arxiv.org/abs/2509.14438
2025
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.